1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext2/xattr.c 4 * 5 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de> 6 * 7 * Fix by Harrison Xing <harrison@mountainviewdata.com>. 8 * Extended attributes for symlinks and special files added per 9 * suggestion of Luka Renko <luka.renko@hermes.si>. 10 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>, 11 * Red Hat Inc. 12 * 13 */ 14 15 /* 16 * Extended attributes are stored on disk blocks allocated outside of 17 * any inode. The i_file_acl field is then made to point to this allocated 18 * block. If all extended attributes of an inode are identical, these 19 * inodes may share the same extended attribute block. Such situations 20 * are automatically detected by keeping a cache of recent attribute block 21 * numbers and hashes over the block's contents in memory. 22 * 23 * 24 * Extended attribute block layout: 25 * 26 * +------------------+ 27 * | header | 28 * | entry 1 | | 29 * | entry 2 | | growing downwards 30 * | entry 3 | v 31 * | four null bytes | 32 * | . . . | 33 * | value 1 | ^ 34 * | value 3 | | growing upwards 35 * | value 2 | | 36 * +------------------+ 37 * 38 * The block header is followed by multiple entry descriptors. These entry 39 * descriptors are variable in size, and aligned to EXT2_XATTR_PAD 40 * byte boundaries. The entry descriptors are sorted by attribute name, 41 * so that two extended attribute blocks can be compared efficiently. 42 * 43 * Attribute values are aligned to the end of the block, stored in 44 * no specific order. They are also padded to EXT2_XATTR_PAD byte 45 * boundaries. No additional gaps are left between them. 46 * 47 * Locking strategy 48 * ---------------- 49 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem. 50 * EA blocks are only changed if they are exclusive to an inode, so 51 * holding xattr_sem also means that nothing but the EA block's reference 52 * count will change. Multiple writers to an EA block are synchronized 53 * by the bh lock. No more than a single bh lock is held at any time 54 * to avoid deadlocks. 55 */ 56 57 #include <linux/buffer_head.h> 58 #include <linux/init.h> 59 #include <linux/slab.h> 60 #include <linux/mbcache.h> 61 #include <linux/quotaops.h> 62 #include <linux/rwsem.h> 63 #include <linux/security.h> 64 #include "ext2.h" 65 #include "xattr.h" 66 #include "acl.h" 67 68 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data)) 69 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr)) 70 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1) 71 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0) 72 73 #ifdef EXT2_XATTR_DEBUG 74 # define ea_idebug(inode, f...) do { \ 75 printk(KERN_DEBUG "inode %s:%ld: ", \ 76 inode->i_sb->s_id, inode->i_ino); \ 77 printk(f); \ 78 printk("\n"); \ 79 } while (0) 80 # define ea_bdebug(bh, f...) do { \ 81 printk(KERN_DEBUG "block %pg:%lu: ", \ 82 bh->b_bdev, (unsigned long) bh->b_blocknr); \ 83 printk(f); \ 84 printk("\n"); \ 85 } while (0) 86 #else 87 # define ea_idebug(f...) 88 # define ea_bdebug(f...) 89 #endif 90 91 static int ext2_xattr_set2(struct inode *, struct buffer_head *, 92 struct ext2_xattr_header *); 93 94 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *); 95 static struct buffer_head *ext2_xattr_cache_find(struct inode *, 96 struct ext2_xattr_header *); 97 static void ext2_xattr_rehash(struct ext2_xattr_header *, 98 struct ext2_xattr_entry *); 99 100 static const struct xattr_handler *ext2_xattr_handler_map[] = { 101 [EXT2_XATTR_INDEX_USER] = &ext2_xattr_user_handler, 102 #ifdef CONFIG_EXT2_FS_POSIX_ACL 103 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler, 104 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler, 105 #endif 106 [EXT2_XATTR_INDEX_TRUSTED] = &ext2_xattr_trusted_handler, 107 #ifdef CONFIG_EXT2_FS_SECURITY 108 [EXT2_XATTR_INDEX_SECURITY] = &ext2_xattr_security_handler, 109 #endif 110 }; 111 112 const struct xattr_handler *ext2_xattr_handlers[] = { 113 &ext2_xattr_user_handler, 114 &ext2_xattr_trusted_handler, 115 #ifdef CONFIG_EXT2_FS_POSIX_ACL 116 &posix_acl_access_xattr_handler, 117 &posix_acl_default_xattr_handler, 118 #endif 119 #ifdef CONFIG_EXT2_FS_SECURITY 120 &ext2_xattr_security_handler, 121 #endif 122 NULL 123 }; 124 125 #define EA_BLOCK_CACHE(inode) (EXT2_SB(inode->i_sb)->s_ea_block_cache) 126 127 static inline const struct xattr_handler * 128 ext2_xattr_handler(int name_index) 129 { 130 const struct xattr_handler *handler = NULL; 131 132 if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map)) 133 handler = ext2_xattr_handler_map[name_index]; 134 return handler; 135 } 136 137 static bool 138 ext2_xattr_header_valid(struct ext2_xattr_header *header) 139 { 140 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 141 header->h_blocks != cpu_to_le32(1)) 142 return false; 143 144 return true; 145 } 146 147 static bool 148 ext2_xattr_entry_valid(struct ext2_xattr_entry *entry, size_t end_offs) 149 { 150 size_t size; 151 152 if (entry->e_value_block != 0) 153 return false; 154 155 size = le32_to_cpu(entry->e_value_size); 156 if (size > end_offs || 157 le16_to_cpu(entry->e_value_offs) + size > end_offs) 158 return false; 159 160 return true; 161 } 162 163 /* 164 * ext2_xattr_get() 165 * 166 * Copy an extended attribute into the buffer 167 * provided, or compute the buffer size required. 168 * Buffer is NULL to compute the size of the buffer required. 169 * 170 * Returns a negative error number on failure, or the number of bytes 171 * used / required on success. 172 */ 173 int 174 ext2_xattr_get(struct inode *inode, int name_index, const char *name, 175 void *buffer, size_t buffer_size) 176 { 177 struct buffer_head *bh = NULL; 178 struct ext2_xattr_entry *entry; 179 size_t name_len, size; 180 char *end; 181 int error; 182 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 183 184 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld", 185 name_index, name, buffer, (long)buffer_size); 186 187 if (name == NULL) 188 return -EINVAL; 189 name_len = strlen(name); 190 if (name_len > 255) 191 return -ERANGE; 192 193 down_read(&EXT2_I(inode)->xattr_sem); 194 error = -ENODATA; 195 if (!EXT2_I(inode)->i_file_acl) 196 goto cleanup; 197 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 198 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 199 error = -EIO; 200 if (!bh) 201 goto cleanup; 202 ea_bdebug(bh, "b_count=%d, refcount=%d", 203 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 204 end = bh->b_data + bh->b_size; 205 if (!ext2_xattr_header_valid(HDR(bh))) { 206 bad_block: 207 ext2_error(inode->i_sb, "ext2_xattr_get", 208 "inode %ld: bad block %d", inode->i_ino, 209 EXT2_I(inode)->i_file_acl); 210 error = -EIO; 211 goto cleanup; 212 } 213 214 /* find named attribute */ 215 entry = FIRST_ENTRY(bh); 216 while (!IS_LAST_ENTRY(entry)) { 217 struct ext2_xattr_entry *next = 218 EXT2_XATTR_NEXT(entry); 219 if ((char *)next >= end) 220 goto bad_block; 221 if (name_index == entry->e_name_index && 222 name_len == entry->e_name_len && 223 memcmp(name, entry->e_name, name_len) == 0) 224 goto found; 225 entry = next; 226 } 227 if (ext2_xattr_cache_insert(ea_block_cache, bh)) 228 ea_idebug(inode, "cache insert failed"); 229 error = -ENODATA; 230 goto cleanup; 231 found: 232 if (!ext2_xattr_entry_valid(entry, inode->i_sb->s_blocksize)) 233 goto bad_block; 234 235 size = le32_to_cpu(entry->e_value_size); 236 if (ext2_xattr_cache_insert(ea_block_cache, bh)) 237 ea_idebug(inode, "cache insert failed"); 238 if (buffer) { 239 error = -ERANGE; 240 if (size > buffer_size) 241 goto cleanup; 242 /* return value of attribute */ 243 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs), 244 size); 245 } 246 error = size; 247 248 cleanup: 249 brelse(bh); 250 up_read(&EXT2_I(inode)->xattr_sem); 251 252 return error; 253 } 254 255 /* 256 * ext2_xattr_list() 257 * 258 * Copy a list of attribute names into the buffer 259 * provided, or compute the buffer size required. 260 * Buffer is NULL to compute the size of the buffer required. 261 * 262 * Returns a negative error number on failure, or the number of bytes 263 * used / required on success. 264 */ 265 static int 266 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size) 267 { 268 struct inode *inode = d_inode(dentry); 269 struct buffer_head *bh = NULL; 270 struct ext2_xattr_entry *entry; 271 char *end; 272 size_t rest = buffer_size; 273 int error; 274 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 275 276 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 277 buffer, (long)buffer_size); 278 279 down_read(&EXT2_I(inode)->xattr_sem); 280 error = 0; 281 if (!EXT2_I(inode)->i_file_acl) 282 goto cleanup; 283 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 284 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 285 error = -EIO; 286 if (!bh) 287 goto cleanup; 288 ea_bdebug(bh, "b_count=%d, refcount=%d", 289 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 290 end = bh->b_data + bh->b_size; 291 if (!ext2_xattr_header_valid(HDR(bh))) { 292 bad_block: 293 ext2_error(inode->i_sb, "ext2_xattr_list", 294 "inode %ld: bad block %d", inode->i_ino, 295 EXT2_I(inode)->i_file_acl); 296 error = -EIO; 297 goto cleanup; 298 } 299 300 /* check the on-disk data structure */ 301 entry = FIRST_ENTRY(bh); 302 while (!IS_LAST_ENTRY(entry)) { 303 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry); 304 305 if ((char *)next >= end) 306 goto bad_block; 307 entry = next; 308 } 309 if (ext2_xattr_cache_insert(ea_block_cache, bh)) 310 ea_idebug(inode, "cache insert failed"); 311 312 /* list the attribute names */ 313 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry); 314 entry = EXT2_XATTR_NEXT(entry)) { 315 const struct xattr_handler *handler = 316 ext2_xattr_handler(entry->e_name_index); 317 318 if (handler && (!handler->list || handler->list(dentry))) { 319 const char *prefix = handler->prefix ?: handler->name; 320 size_t prefix_len = strlen(prefix); 321 size_t size = prefix_len + entry->e_name_len + 1; 322 323 if (buffer) { 324 if (size > rest) { 325 error = -ERANGE; 326 goto cleanup; 327 } 328 memcpy(buffer, prefix, prefix_len); 329 buffer += prefix_len; 330 memcpy(buffer, entry->e_name, entry->e_name_len); 331 buffer += entry->e_name_len; 332 *buffer++ = 0; 333 } 334 rest -= size; 335 } 336 } 337 error = buffer_size - rest; /* total size */ 338 339 cleanup: 340 brelse(bh); 341 up_read(&EXT2_I(inode)->xattr_sem); 342 343 return error; 344 } 345 346 /* 347 * Inode operation listxattr() 348 * 349 * d_inode(dentry)->i_mutex: don't care 350 */ 351 ssize_t 352 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size) 353 { 354 return ext2_xattr_list(dentry, buffer, size); 355 } 356 357 /* 358 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is 359 * not set, set it. 360 */ 361 static void ext2_xattr_update_super_block(struct super_block *sb) 362 { 363 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR)) 364 return; 365 366 spin_lock(&EXT2_SB(sb)->s_lock); 367 ext2_update_dynamic_rev(sb); 368 EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR); 369 spin_unlock(&EXT2_SB(sb)->s_lock); 370 mark_buffer_dirty(EXT2_SB(sb)->s_sbh); 371 } 372 373 /* 374 * ext2_xattr_set() 375 * 376 * Create, replace or remove an extended attribute for this inode. Value 377 * is NULL to remove an existing extended attribute, and non-NULL to 378 * either replace an existing extended attribute, or create a new extended 379 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 380 * specify that an extended attribute must exist and must not exist 381 * previous to the call, respectively. 382 * 383 * Returns 0, or a negative error number on failure. 384 */ 385 int 386 ext2_xattr_set(struct inode *inode, int name_index, const char *name, 387 const void *value, size_t value_len, int flags) 388 { 389 struct super_block *sb = inode->i_sb; 390 struct buffer_head *bh = NULL; 391 struct ext2_xattr_header *header = NULL; 392 struct ext2_xattr_entry *here, *last; 393 size_t name_len, free, min_offs = sb->s_blocksize; 394 int not_found = 1, error; 395 char *end; 396 397 /* 398 * header -- Points either into bh, or to a temporarily 399 * allocated buffer. 400 * here -- The named entry found, or the place for inserting, within 401 * the block pointed to by header. 402 * last -- Points right after the last named entry within the block 403 * pointed to by header. 404 * min_offs -- The offset of the first value (values are aligned 405 * towards the end of the block). 406 * end -- Points right after the block pointed to by header. 407 */ 408 409 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 410 name_index, name, value, (long)value_len); 411 412 if (value == NULL) 413 value_len = 0; 414 if (name == NULL) 415 return -EINVAL; 416 name_len = strlen(name); 417 if (name_len > 255 || value_len > sb->s_blocksize) 418 return -ERANGE; 419 down_write(&EXT2_I(inode)->xattr_sem); 420 if (EXT2_I(inode)->i_file_acl) { 421 /* The inode already has an extended attribute block. */ 422 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl); 423 error = -EIO; 424 if (!bh) 425 goto cleanup; 426 ea_bdebug(bh, "b_count=%d, refcount=%d", 427 atomic_read(&(bh->b_count)), 428 le32_to_cpu(HDR(bh)->h_refcount)); 429 header = HDR(bh); 430 end = bh->b_data + bh->b_size; 431 if (!ext2_xattr_header_valid(header)) { 432 bad_block: 433 ext2_error(sb, "ext2_xattr_set", 434 "inode %ld: bad block %d", inode->i_ino, 435 EXT2_I(inode)->i_file_acl); 436 error = -EIO; 437 goto cleanup; 438 } 439 /* 440 * Find the named attribute. If not found, 'here' will point 441 * to entry where the new attribute should be inserted to 442 * maintain sorting. 443 */ 444 last = FIRST_ENTRY(bh); 445 while (!IS_LAST_ENTRY(last)) { 446 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last); 447 if ((char *)next >= end) 448 goto bad_block; 449 if (!last->e_value_block && last->e_value_size) { 450 size_t offs = le16_to_cpu(last->e_value_offs); 451 if (offs < min_offs) 452 min_offs = offs; 453 } 454 if (not_found > 0) { 455 not_found = name_index - last->e_name_index; 456 if (!not_found) 457 not_found = name_len - last->e_name_len; 458 if (!not_found) { 459 not_found = memcmp(name, last->e_name, 460 name_len); 461 } 462 if (not_found <= 0) 463 here = last; 464 } 465 last = next; 466 } 467 if (not_found > 0) 468 here = last; 469 470 /* Check whether we have enough space left. */ 471 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32); 472 } else { 473 /* We will use a new extended attribute block. */ 474 free = sb->s_blocksize - 475 sizeof(struct ext2_xattr_header) - sizeof(__u32); 476 here = last = NULL; /* avoid gcc uninitialized warning. */ 477 } 478 479 if (not_found) { 480 /* Request to remove a nonexistent attribute? */ 481 error = -ENODATA; 482 if (flags & XATTR_REPLACE) 483 goto cleanup; 484 error = 0; 485 if (value == NULL) 486 goto cleanup; 487 } else { 488 /* Request to create an existing attribute? */ 489 error = -EEXIST; 490 if (flags & XATTR_CREATE) 491 goto cleanup; 492 if (!here->e_value_block && here->e_value_size) { 493 if (!ext2_xattr_entry_valid(here, sb->s_blocksize)) 494 goto bad_block; 495 free += EXT2_XATTR_SIZE( 496 le32_to_cpu(here->e_value_size)); 497 } 498 free += EXT2_XATTR_LEN(name_len); 499 } 500 error = -ENOSPC; 501 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len)) 502 goto cleanup; 503 504 /* Here we know that we can set the new attribute. */ 505 506 if (header) { 507 /* assert(header == HDR(bh)); */ 508 lock_buffer(bh); 509 if (header->h_refcount == cpu_to_le32(1)) { 510 __u32 hash = le32_to_cpu(header->h_hash); 511 512 ea_bdebug(bh, "modifying in-place"); 513 /* 514 * This must happen under buffer lock for 515 * ext2_xattr_set2() to reliably detect modified block 516 */ 517 mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash, 518 bh->b_blocknr); 519 520 /* keep the buffer locked while modifying it. */ 521 } else { 522 int offset; 523 524 unlock_buffer(bh); 525 ea_bdebug(bh, "cloning"); 526 header = kmalloc(bh->b_size, GFP_KERNEL); 527 error = -ENOMEM; 528 if (header == NULL) 529 goto cleanup; 530 memcpy(header, HDR(bh), bh->b_size); 531 header->h_refcount = cpu_to_le32(1); 532 533 offset = (char *)here - bh->b_data; 534 here = ENTRY((char *)header + offset); 535 offset = (char *)last - bh->b_data; 536 last = ENTRY((char *)header + offset); 537 } 538 } else { 539 /* Allocate a buffer where we construct the new block. */ 540 header = kzalloc(sb->s_blocksize, GFP_KERNEL); 541 error = -ENOMEM; 542 if (header == NULL) 543 goto cleanup; 544 end = (char *)header + sb->s_blocksize; 545 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC); 546 header->h_blocks = header->h_refcount = cpu_to_le32(1); 547 last = here = ENTRY(header+1); 548 } 549 550 /* Iff we are modifying the block in-place, bh is locked here. */ 551 552 if (not_found) { 553 /* Insert the new name. */ 554 size_t size = EXT2_XATTR_LEN(name_len); 555 size_t rest = (char *)last - (char *)here; 556 memmove((char *)here + size, here, rest); 557 memset(here, 0, size); 558 here->e_name_index = name_index; 559 here->e_name_len = name_len; 560 memcpy(here->e_name, name, name_len); 561 } else { 562 if (!here->e_value_block && here->e_value_size) { 563 char *first_val = (char *)header + min_offs; 564 size_t offs = le16_to_cpu(here->e_value_offs); 565 char *val = (char *)header + offs; 566 size_t size = EXT2_XATTR_SIZE( 567 le32_to_cpu(here->e_value_size)); 568 569 if (size == EXT2_XATTR_SIZE(value_len)) { 570 /* The old and the new value have the same 571 size. Just replace. */ 572 here->e_value_size = cpu_to_le32(value_len); 573 memset(val + size - EXT2_XATTR_PAD, 0, 574 EXT2_XATTR_PAD); /* Clear pad bytes. */ 575 memcpy(val, value, value_len); 576 goto skip_replace; 577 } 578 579 /* Remove the old value. */ 580 memmove(first_val + size, first_val, val - first_val); 581 memset(first_val, 0, size); 582 here->e_value_offs = 0; 583 min_offs += size; 584 585 /* Adjust all value offsets. */ 586 last = ENTRY(header+1); 587 while (!IS_LAST_ENTRY(last)) { 588 size_t o = le16_to_cpu(last->e_value_offs); 589 if (!last->e_value_block && o < offs) 590 last->e_value_offs = 591 cpu_to_le16(o + size); 592 last = EXT2_XATTR_NEXT(last); 593 } 594 } 595 if (value == NULL) { 596 /* Remove the old name. */ 597 size_t size = EXT2_XATTR_LEN(name_len); 598 last = ENTRY((char *)last - size); 599 memmove(here, (char*)here + size, 600 (char*)last - (char*)here); 601 memset(last, 0, size); 602 } 603 } 604 605 if (value != NULL) { 606 /* Insert the new value. */ 607 here->e_value_size = cpu_to_le32(value_len); 608 if (value_len) { 609 size_t size = EXT2_XATTR_SIZE(value_len); 610 char *val = (char *)header + min_offs - size; 611 here->e_value_offs = 612 cpu_to_le16((char *)val - (char *)header); 613 memset(val + size - EXT2_XATTR_PAD, 0, 614 EXT2_XATTR_PAD); /* Clear the pad bytes. */ 615 memcpy(val, value, value_len); 616 } 617 } 618 619 skip_replace: 620 if (IS_LAST_ENTRY(ENTRY(header+1))) { 621 /* This block is now empty. */ 622 if (bh && header == HDR(bh)) 623 unlock_buffer(bh); /* we were modifying in-place. */ 624 error = ext2_xattr_set2(inode, bh, NULL); 625 } else { 626 ext2_xattr_rehash(header, here); 627 if (bh && header == HDR(bh)) 628 unlock_buffer(bh); /* we were modifying in-place. */ 629 error = ext2_xattr_set2(inode, bh, header); 630 } 631 632 cleanup: 633 if (!(bh && header == HDR(bh))) 634 kfree(header); 635 brelse(bh); 636 up_write(&EXT2_I(inode)->xattr_sem); 637 638 return error; 639 } 640 641 /* 642 * Second half of ext2_xattr_set(): Update the file system. 643 */ 644 static int 645 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh, 646 struct ext2_xattr_header *header) 647 { 648 struct super_block *sb = inode->i_sb; 649 struct buffer_head *new_bh = NULL; 650 int error; 651 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 652 653 if (header) { 654 new_bh = ext2_xattr_cache_find(inode, header); 655 if (new_bh) { 656 /* We found an identical block in the cache. */ 657 if (new_bh == old_bh) { 658 ea_bdebug(new_bh, "keeping this block"); 659 } else { 660 /* The old block is released after updating 661 the inode. */ 662 ea_bdebug(new_bh, "reusing block"); 663 664 error = dquot_alloc_block(inode, 1); 665 if (error) { 666 unlock_buffer(new_bh); 667 goto cleanup; 668 } 669 le32_add_cpu(&HDR(new_bh)->h_refcount, 1); 670 ea_bdebug(new_bh, "refcount now=%d", 671 le32_to_cpu(HDR(new_bh)->h_refcount)); 672 } 673 unlock_buffer(new_bh); 674 } else if (old_bh && header == HDR(old_bh)) { 675 /* Keep this block. No need to lock the block as we 676 don't need to change the reference count. */ 677 new_bh = old_bh; 678 get_bh(new_bh); 679 ext2_xattr_cache_insert(ea_block_cache, new_bh); 680 } else { 681 /* We need to allocate a new block */ 682 ext2_fsblk_t goal = ext2_group_first_block_no(sb, 683 EXT2_I(inode)->i_block_group); 684 int block = ext2_new_block(inode, goal, &error); 685 if (error) 686 goto cleanup; 687 ea_idebug(inode, "creating block %d", block); 688 689 new_bh = sb_getblk(sb, block); 690 if (unlikely(!new_bh)) { 691 ext2_free_blocks(inode, block, 1); 692 mark_inode_dirty(inode); 693 error = -ENOMEM; 694 goto cleanup; 695 } 696 lock_buffer(new_bh); 697 memcpy(new_bh->b_data, header, new_bh->b_size); 698 set_buffer_uptodate(new_bh); 699 unlock_buffer(new_bh); 700 ext2_xattr_cache_insert(ea_block_cache, new_bh); 701 702 ext2_xattr_update_super_block(sb); 703 } 704 mark_buffer_dirty(new_bh); 705 if (IS_SYNC(inode)) { 706 sync_dirty_buffer(new_bh); 707 error = -EIO; 708 if (buffer_req(new_bh) && !buffer_uptodate(new_bh)) 709 goto cleanup; 710 } 711 } 712 713 /* Update the inode. */ 714 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 715 inode->i_ctime = current_time(inode); 716 if (IS_SYNC(inode)) { 717 error = sync_inode_metadata(inode, 1); 718 /* In case sync failed due to ENOSPC the inode was actually 719 * written (only some dirty data were not) so we just proceed 720 * as if nothing happened and cleanup the unused block */ 721 if (error && error != -ENOSPC) { 722 if (new_bh && new_bh != old_bh) { 723 dquot_free_block_nodirty(inode, 1); 724 mark_inode_dirty(inode); 725 } 726 goto cleanup; 727 } 728 } else 729 mark_inode_dirty(inode); 730 731 error = 0; 732 if (old_bh && old_bh != new_bh) { 733 /* 734 * If there was an old block and we are no longer using it, 735 * release the old block. 736 */ 737 lock_buffer(old_bh); 738 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) { 739 __u32 hash = le32_to_cpu(HDR(old_bh)->h_hash); 740 741 /* 742 * This must happen under buffer lock for 743 * ext2_xattr_set2() to reliably detect freed block 744 */ 745 mb_cache_entry_delete(ea_block_cache, hash, 746 old_bh->b_blocknr); 747 /* Free the old block. */ 748 ea_bdebug(old_bh, "freeing"); 749 ext2_free_blocks(inode, old_bh->b_blocknr, 1); 750 mark_inode_dirty(inode); 751 /* We let our caller release old_bh, so we 752 * need to duplicate the buffer before. */ 753 get_bh(old_bh); 754 bforget(old_bh); 755 } else { 756 /* Decrement the refcount only. */ 757 le32_add_cpu(&HDR(old_bh)->h_refcount, -1); 758 dquot_free_block_nodirty(inode, 1); 759 mark_inode_dirty(inode); 760 mark_buffer_dirty(old_bh); 761 ea_bdebug(old_bh, "refcount now=%d", 762 le32_to_cpu(HDR(old_bh)->h_refcount)); 763 } 764 unlock_buffer(old_bh); 765 } 766 767 cleanup: 768 brelse(new_bh); 769 770 return error; 771 } 772 773 /* 774 * ext2_xattr_delete_inode() 775 * 776 * Free extended attribute resources associated with this inode. This 777 * is called immediately before an inode is freed. 778 */ 779 void 780 ext2_xattr_delete_inode(struct inode *inode) 781 { 782 struct buffer_head *bh = NULL; 783 struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb); 784 785 down_write(&EXT2_I(inode)->xattr_sem); 786 if (!EXT2_I(inode)->i_file_acl) 787 goto cleanup; 788 789 if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) { 790 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 791 "inode %ld: xattr block %d is out of data blocks range", 792 inode->i_ino, EXT2_I(inode)->i_file_acl); 793 goto cleanup; 794 } 795 796 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 797 if (!bh) { 798 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 799 "inode %ld: block %d read error", inode->i_ino, 800 EXT2_I(inode)->i_file_acl); 801 goto cleanup; 802 } 803 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count))); 804 if (!ext2_xattr_header_valid(HDR(bh))) { 805 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 806 "inode %ld: bad block %d", inode->i_ino, 807 EXT2_I(inode)->i_file_acl); 808 goto cleanup; 809 } 810 lock_buffer(bh); 811 if (HDR(bh)->h_refcount == cpu_to_le32(1)) { 812 __u32 hash = le32_to_cpu(HDR(bh)->h_hash); 813 814 /* 815 * This must happen under buffer lock for ext2_xattr_set2() to 816 * reliably detect freed block 817 */ 818 mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash, 819 bh->b_blocknr); 820 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1); 821 get_bh(bh); 822 bforget(bh); 823 unlock_buffer(bh); 824 } else { 825 le32_add_cpu(&HDR(bh)->h_refcount, -1); 826 ea_bdebug(bh, "refcount now=%d", 827 le32_to_cpu(HDR(bh)->h_refcount)); 828 unlock_buffer(bh); 829 mark_buffer_dirty(bh); 830 if (IS_SYNC(inode)) 831 sync_dirty_buffer(bh); 832 dquot_free_block_nodirty(inode, 1); 833 } 834 EXT2_I(inode)->i_file_acl = 0; 835 836 cleanup: 837 brelse(bh); 838 up_write(&EXT2_I(inode)->xattr_sem); 839 } 840 841 /* 842 * ext2_xattr_cache_insert() 843 * 844 * Create a new entry in the extended attribute cache, and insert 845 * it unless such an entry is already in the cache. 846 * 847 * Returns 0, or a negative error number on failure. 848 */ 849 static int 850 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh) 851 { 852 __u32 hash = le32_to_cpu(HDR(bh)->h_hash); 853 int error; 854 855 error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr, 856 true); 857 if (error) { 858 if (error == -EBUSY) { 859 ea_bdebug(bh, "already in cache (%d cache entries)", 860 atomic_read(&ext2_xattr_cache->c_entry_count)); 861 error = 0; 862 } 863 } else 864 ea_bdebug(bh, "inserting [%x]", (int)hash); 865 return error; 866 } 867 868 /* 869 * ext2_xattr_cmp() 870 * 871 * Compare two extended attribute blocks for equality. 872 * 873 * Returns 0 if the blocks are equal, 1 if they differ, and 874 * a negative error number on errors. 875 */ 876 static int 877 ext2_xattr_cmp(struct ext2_xattr_header *header1, 878 struct ext2_xattr_header *header2) 879 { 880 struct ext2_xattr_entry *entry1, *entry2; 881 882 entry1 = ENTRY(header1+1); 883 entry2 = ENTRY(header2+1); 884 while (!IS_LAST_ENTRY(entry1)) { 885 if (IS_LAST_ENTRY(entry2)) 886 return 1; 887 if (entry1->e_hash != entry2->e_hash || 888 entry1->e_name_index != entry2->e_name_index || 889 entry1->e_name_len != entry2->e_name_len || 890 entry1->e_value_size != entry2->e_value_size || 891 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 892 return 1; 893 if (entry1->e_value_block != 0 || entry2->e_value_block != 0) 894 return -EIO; 895 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 896 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 897 le32_to_cpu(entry1->e_value_size))) 898 return 1; 899 900 entry1 = EXT2_XATTR_NEXT(entry1); 901 entry2 = EXT2_XATTR_NEXT(entry2); 902 } 903 if (!IS_LAST_ENTRY(entry2)) 904 return 1; 905 return 0; 906 } 907 908 /* 909 * ext2_xattr_cache_find() 910 * 911 * Find an identical extended attribute block. 912 * 913 * Returns a locked buffer head to the block found, or NULL if such 914 * a block was not found or an error occurred. 915 */ 916 static struct buffer_head * 917 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header) 918 { 919 __u32 hash = le32_to_cpu(header->h_hash); 920 struct mb_cache_entry *ce; 921 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 922 923 if (!header->h_hash) 924 return NULL; /* never share */ 925 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 926 again: 927 ce = mb_cache_entry_find_first(ea_block_cache, hash); 928 while (ce) { 929 struct buffer_head *bh; 930 931 bh = sb_bread(inode->i_sb, ce->e_value); 932 if (!bh) { 933 ext2_error(inode->i_sb, "ext2_xattr_cache_find", 934 "inode %ld: block %ld read error", 935 inode->i_ino, (unsigned long) ce->e_value); 936 } else { 937 lock_buffer(bh); 938 /* 939 * We have to be careful about races with freeing or 940 * rehashing of xattr block. Once we hold buffer lock 941 * xattr block's state is stable so we can check 942 * whether the block got freed / rehashed or not. 943 * Since we unhash mbcache entry under buffer lock when 944 * freeing / rehashing xattr block, checking whether 945 * entry is still hashed is reliable. 946 */ 947 if (hlist_bl_unhashed(&ce->e_hash_list)) { 948 mb_cache_entry_put(ea_block_cache, ce); 949 unlock_buffer(bh); 950 brelse(bh); 951 goto again; 952 } else if (le32_to_cpu(HDR(bh)->h_refcount) > 953 EXT2_XATTR_REFCOUNT_MAX) { 954 ea_idebug(inode, "block %ld refcount %d>%d", 955 (unsigned long) ce->e_value, 956 le32_to_cpu(HDR(bh)->h_refcount), 957 EXT2_XATTR_REFCOUNT_MAX); 958 } else if (!ext2_xattr_cmp(header, HDR(bh))) { 959 ea_bdebug(bh, "b_count=%d", 960 atomic_read(&(bh->b_count))); 961 mb_cache_entry_touch(ea_block_cache, ce); 962 mb_cache_entry_put(ea_block_cache, ce); 963 return bh; 964 } 965 unlock_buffer(bh); 966 brelse(bh); 967 } 968 ce = mb_cache_entry_find_next(ea_block_cache, ce); 969 } 970 return NULL; 971 } 972 973 #define NAME_HASH_SHIFT 5 974 #define VALUE_HASH_SHIFT 16 975 976 /* 977 * ext2_xattr_hash_entry() 978 * 979 * Compute the hash of an extended attribute. 980 */ 981 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header, 982 struct ext2_xattr_entry *entry) 983 { 984 __u32 hash = 0; 985 char *name = entry->e_name; 986 int n; 987 988 for (n=0; n < entry->e_name_len; n++) { 989 hash = (hash << NAME_HASH_SHIFT) ^ 990 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 991 *name++; 992 } 993 994 if (entry->e_value_block == 0 && entry->e_value_size != 0) { 995 __le32 *value = (__le32 *)((char *)header + 996 le16_to_cpu(entry->e_value_offs)); 997 for (n = (le32_to_cpu(entry->e_value_size) + 998 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) { 999 hash = (hash << VALUE_HASH_SHIFT) ^ 1000 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 1001 le32_to_cpu(*value++); 1002 } 1003 } 1004 entry->e_hash = cpu_to_le32(hash); 1005 } 1006 1007 #undef NAME_HASH_SHIFT 1008 #undef VALUE_HASH_SHIFT 1009 1010 #define BLOCK_HASH_SHIFT 16 1011 1012 /* 1013 * ext2_xattr_rehash() 1014 * 1015 * Re-compute the extended attribute hash value after an entry has changed. 1016 */ 1017 static void ext2_xattr_rehash(struct ext2_xattr_header *header, 1018 struct ext2_xattr_entry *entry) 1019 { 1020 struct ext2_xattr_entry *here; 1021 __u32 hash = 0; 1022 1023 ext2_xattr_hash_entry(header, entry); 1024 here = ENTRY(header+1); 1025 while (!IS_LAST_ENTRY(here)) { 1026 if (!here->e_hash) { 1027 /* Block is not shared if an entry's hash value == 0 */ 1028 hash = 0; 1029 break; 1030 } 1031 hash = (hash << BLOCK_HASH_SHIFT) ^ 1032 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 1033 le32_to_cpu(here->e_hash); 1034 here = EXT2_XATTR_NEXT(here); 1035 } 1036 header->h_hash = cpu_to_le32(hash); 1037 } 1038 1039 #undef BLOCK_HASH_SHIFT 1040 1041 #define HASH_BUCKET_BITS 10 1042 1043 struct mb_cache *ext2_xattr_create_cache(void) 1044 { 1045 return mb_cache_create(HASH_BUCKET_BITS); 1046 } 1047 1048 void ext2_xattr_destroy_cache(struct mb_cache *cache) 1049 { 1050 if (cache) 1051 mb_cache_destroy(cache); 1052 } 1053